By: Daniel Cooper, PGY-3
I furrowed my brow at the red letters stating that my artificial intelligence (AI) clinical tool was experiencing high volume. Surely my simple query — “Should the emergency department have started this patient on vancomycin and piperacillin-tazobactam?” — should be enough for the iota of computing power it needs to process.
Searching another AI bot, at least I got an answer this time. Maybe I was not asking the correct kinds of questions, because I got a result I was not in the mood for: It depends on the patient. Depends how? Give me a deep consult and keep it between the guidelines and evidence, I tell my trusty sidekick. How can I have this decision made for me? What did people do before we had fancy technology to help make our clinical decisions? Clinical intuition? Gestalt? A colleague? Tinsley Harrison’s textbook?
Sighing, I trudged down to the ICU to obtain the rest of my history. It was a frail patient with a partial laryngectomy and a tracheal stoma that he had to cover to speak. Leaning in close to make out his garbled words, I heard all about the events leading up to his hospitalization, including his weight loss and low blood pressure that landed him in the ICU, but most importantly, that he had lost his phone. Locating it next to him, I moved it closer, taped his phone charger to the bed so it would not be lost, and wished him a speedy recovery. In response to these small gestures, through his secretion-laden stoma, he uttered, “You’re a good man, doc.”
My scribbling with pen and paper stopped, and the furrow in my brow lightened. Oculus 3000 would not have paused, instead continuing its ambient omniscience, generating chains of words in an iterative process pleasing to human eyes. What he said, though, was impactful, reminding me of the importance of staying at the bedside; our presence is irreplaceable.
Minutes after leaving the bedside, a nurse paged me, saying that something about her patient “just didn’t feel right.” This is something you take seriously, especially from a seasoned nurse. That patient ended up intubated for altered mental status, and we broadened his coverage to treat the meningitis we suspected. Could TimecutterPRO have accurately predicted a decompensation like that, or was her gestalt what saved his life? If we had entered all his vitals and clinical appearance, I cannot help but wonder whether artificial intelligence would have been smart enough to have a hunch.
As someone whose brain appreciates the decrease in cognitive burden associated with searching treatment strategies, I am forever worried my brain will turn to mush despite attempts to reunite us with patients. Recognizing that AI is inevitable, and making no mistake that humans, myself included, are fallible and error-prone, there is still nothing more important than embracing fewer overthinking caps. That said, someone has to be the person to tell another they are dying, to hold the hand of the agonal breather, and to choose to start the broad-spectrum antibiotics. Do you really want ProfitMaximizerGPT to tell a mom about her leukemia?